0016-add-model-quantizations.patch

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: jmorganca <jmorganca@gmail.com>
Date: Tue, 8 Apr 2025 20:39:32 -0700
Subject: [PATCH] add model quantizations

a temporary patch to add model quantization for
models not supported in llama.cpp
---
 src/llama-arch.cpp  | 17 +++++++++++++++++
 src/llama-arch.h    |  1 +
 src/llama-model.cpp |  2 ++
 src/llama-quant.cpp |  4 ++++
 4 files changed, 24 insertions(+)

diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index eb7b5325..df42d1a5 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -74,6 +74,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
     { LLM_ARCH_PLM,              "plm"              },
     { LLM_ARCH_BAILINGMOE,       "bailingmoe"       },
+    { LLM_ARCH_MISTRAL3,         "mistral3"         },
     { LLM_ARCH_UNKNOWN,          "(unknown)"        },
 };
 
@@ -1606,6 +1607,22 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
         },
     },
+    {
+        LLM_ARCH_MISTRAL3,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,  "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM, "output_norm" },
+            { LLM_TENSOR_ATTN_NORM,   "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_Q,      "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_K,      "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_V,      "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,    "blk.%d.attn_output" },
+            { LLM_TENSOR_FFN_NORM,    "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_GATE,    "blk.%d.ffn_gate" },
+            { LLM_TENSOR_FFN_UP,      "blk.%d.ffn_up" },
+            { LLM_TENSOR_FFN_DOWN,    "blk.%d.ffn_down" },
+        }
+    },
     {
         LLM_ARCH_UNKNOWN,
         {
diff --git a/src/llama-arch.h b/src/llama-arch.h
index bc8a4f0b..bda9d071 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -76,6 +76,7 @@ enum llm_arch {
     LLM_ARCH_CHAMELEON,
     LLM_ARCH_SOLAR,
     LLM_ARCH_WAVTOKENIZER_DEC,
+    LLM_ARCH_MISTRAL3,
     LLM_ARCH_PLM,
     LLM_ARCH_BAILINGMOE,
     LLM_ARCH_UNKNOWN,
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index 9d099f11..ef70486d 100644
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -1437,6 +1437,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                     default: type = LLM_TYPE_UNKNOWN;
                 }
             } break;
+        case LLM_ARCH_MISTRAL3: break;
         default: throw std::runtime_error("unsupported model architecture");
     }
 
@@ -13751,6 +13752,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
         case LLM_ARCH_CHAMELEON:
         case LLM_ARCH_SOLAR:
         case LLM_ARCH_BAILINGMOE:
+        case LLM_ARCH_MISTRAL3:
             return LLAMA_ROPE_TYPE_NORM;
 
         // the pairs of head values are offset by n_rot/2
diff --git a/src/llama-quant.cpp b/src/llama-quant.cpp
index 223e1f3f..8ae6dde8 100644
--- a/src/llama-quant.cpp
+++ b/src/llama-quant.cpp
@@ -744,6 +744,10 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
         // This used to be a regex, but <regex> has an extreme cost to compile times.
         bool quantize = name.rfind("weight") == name.size() - 6; // ends with 'weight'?
 
+        // don't quantize vision stuff
+        quantize &= name.find("v.") == std::string::npos;
+        quantize &= name.find("mm.") == std::string::npos;
+
         // quantize only 2D and 3D tensors (experts)
         quantize &= (ggml_n_dims(tensor) >= 2);